Joint generative and discriminative models for spoken language understanding

Marco Dinarelli, Alessandro Moschitti, Giuseppe Riccardi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

Spoken Language Understanding aims at mapping a naturallanguage spoken sentence into a semantic representation. In the last decade two main approaches have been pursued: generative and discriminative models. The former is more robust to overfitting whereas the latter is more robust to many irrelevant features. Additionally, the way in which these approaches encode prior knowledge is very different and their relative performance changes based on the task. In this paper we describe a training framework where both models are used: a generative model produces a list of ranked hypotheses whereas a discriminative model, depending on string kernels and Support Vector Machines, re-ranks such list. We tested such approach on a new corpus produced in the European LUNA project. The results show a large improvement on the state-of-the-art in concept segmentation and labeling.

Original languageEnglish
Title of host publication2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings
Pages61-64
Number of pages4
DOIs
Publication statusPublished - 1 Dec 2008
Externally publishedYes
Event2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Goa, India
Duration: 15 Dec 200819 Dec 2008

Other

Other2008 IEEE Workshop on Spoken Language Technology, SLT 2008
CountryIndia
CityGoa
Period15/12/0819/12/08

Fingerprint

spoken language
Labeling
Support vector machines
Semantics
semantics
Generative
Spoken Language Understanding
knowledge
performance

Keywords

  • Finite state transducers
  • Generative and discriminative models
  • Kernel methods
  • Spoken language understanding
  • Stochastic language models

ASJC Scopus subject areas

  • Language and Linguistics
  • Software
  • Electrical and Electronic Engineering
  • Communication

Cite this

Dinarelli, M., Moschitti, A., & Riccardi, G. (2008). Joint generative and discriminative models for spoken language understanding. In 2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings (pp. 61-64). [4777840] https://doi.org/10.1109/SLT.2008.4777840

Joint generative and discriminative models for spoken language understanding. / Dinarelli, Marco; Moschitti, Alessandro; Riccardi, Giuseppe.

2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings. 2008. p. 61-64 4777840.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Dinarelli, M, Moschitti, A & Riccardi, G 2008, Joint generative and discriminative models for spoken language understanding. in 2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings., 4777840, pp. 61-64, 2008 IEEE Workshop on Spoken Language Technology, SLT 2008, Goa, India, 15/12/08. https://doi.org/10.1109/SLT.2008.4777840
Dinarelli M, Moschitti A, Riccardi G. Joint generative and discriminative models for spoken language understanding. In 2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings. 2008. p. 61-64. 4777840 https://doi.org/10.1109/SLT.2008.4777840
Dinarelli, Marco ; Moschitti, Alessandro ; Riccardi, Giuseppe. / Joint generative and discriminative models for spoken language understanding. 2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings. 2008. pp. 61-64
@inproceedings{c03e8f95aaf044e6b4ad8df42148515f,
title = "Joint generative and discriminative models for spoken language understanding",
abstract = "Spoken Language Understanding aims at mapping a naturallanguage spoken sentence into a semantic representation. In the last decade two main approaches have been pursued: generative and discriminative models. The former is more robust to overfitting whereas the latter is more robust to many irrelevant features. Additionally, the way in which these approaches encode prior knowledge is very different and their relative performance changes based on the task. In this paper we describe a training framework where both models are used: a generative model produces a list of ranked hypotheses whereas a discriminative model, depending on string kernels and Support Vector Machines, re-ranks such list. We tested such approach on a new corpus produced in the European LUNA project. The results show a large improvement on the state-of-the-art in concept segmentation and labeling.",
keywords = "Finite state transducers, Generative and discriminative models, Kernel methods, Spoken language understanding, Stochastic language models",
author = "Marco Dinarelli and Alessandro Moschitti and Giuseppe Riccardi",
year = "2008",
month = "12",
day = "1",
doi = "10.1109/SLT.2008.4777840",
language = "English",
isbn = "9781424434725",
pages = "61--64",
booktitle = "2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings",

}

TY - GEN

T1 - Joint generative and discriminative models for spoken language understanding

AU - Dinarelli, Marco

AU - Moschitti, Alessandro

AU - Riccardi, Giuseppe

PY - 2008/12/1

Y1 - 2008/12/1

N2 - Spoken Language Understanding aims at mapping a naturallanguage spoken sentence into a semantic representation. In the last decade two main approaches have been pursued: generative and discriminative models. The former is more robust to overfitting whereas the latter is more robust to many irrelevant features. Additionally, the way in which these approaches encode prior knowledge is very different and their relative performance changes based on the task. In this paper we describe a training framework where both models are used: a generative model produces a list of ranked hypotheses whereas a discriminative model, depending on string kernels and Support Vector Machines, re-ranks such list. We tested such approach on a new corpus produced in the European LUNA project. The results show a large improvement on the state-of-the-art in concept segmentation and labeling.

AB - Spoken Language Understanding aims at mapping a naturallanguage spoken sentence into a semantic representation. In the last decade two main approaches have been pursued: generative and discriminative models. The former is more robust to overfitting whereas the latter is more robust to many irrelevant features. Additionally, the way in which these approaches encode prior knowledge is very different and their relative performance changes based on the task. In this paper we describe a training framework where both models are used: a generative model produces a list of ranked hypotheses whereas a discriminative model, depending on string kernels and Support Vector Machines, re-ranks such list. We tested such approach on a new corpus produced in the European LUNA project. The results show a large improvement on the state-of-the-art in concept segmentation and labeling.

KW - Finite state transducers

KW - Generative and discriminative models

KW - Kernel methods

KW - Spoken language understanding

KW - Stochastic language models

UR - http://www.scopus.com/inward/record.url?scp=67649512443&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=67649512443&partnerID=8YFLogxK

U2 - 10.1109/SLT.2008.4777840

DO - 10.1109/SLT.2008.4777840

M3 - Conference contribution

AN - SCOPUS:67649512443

SN - 9781424434725

SP - 61

EP - 64

BT - 2008 IEEE Workshop on Spoken Language Technology, SLT 2008 - Proceedings

ER -